Icarus 251 (2015) 112–127

Contents lists available at ScienceDirect

Icarus

journal homepage: www.elsevier .com/ locate/ icarus

The origin, evolution, and trajectory of large dust storms on Mars duringMars years 24–30 (1999–2011)

0019-1035/$ - see front matter � 2013 Elsevier Inc. All rights reserved.http://dx.doi.org/10.1016/j.icarus.2013.10.033

⇑ Corresponding author. Address: MS 50, 60 Garden Street, Cambridge, MA02138, USA.

E-mail address: hwang@cfa.harvard.edu (H. Wang).

Huiqun Wang a,⇑, Mark I. Richardson b

a Smithsonian Astrophysical Observatory, Cambridge, MA 02138, USAb Ashima Research, Pasadena, CA 91101, USA

a r t i c l e i n f o

Article history:Available online 14 November 2013

Keywords:MarsMars, atmosphereImage processing

a b s t r a c t

Mars Daily Global Maps (MDGM) derived from the Mars Global Surveyor (MGS) Mars Orbiter Camera(MOC) and Mars Reconnaissance Orbiter (MRO) Mars Color Imager (MARCI) are used to study the distri-bution and evolution of large dust storms over the period from Mars years 24–30 (1999–2001). Largestorms are defined here as discrete dust events visible in image sequences extending over at least 5 sols(Mars days) and where the dust covers areas beyond the origination region. A total of 65 large duststorms meeting these criteria are identified during the observational period and all are observed duringthe Ls = 135–30� seasonal window. Dust storms originating in the northern and southern hemispheresappear to form two distinct families. All but two of the storms originating in the northern hemisphereare observed in two seasonal windows at Ls = 180–240� and Ls = 305–350�; while all but two of thoseoriginating in the southern hemisphere are observed during Ls = 135–245�.

None of the large dust storms originating in the northern hemisphere are observed to develop to globalscale, but some of them develop into large regional storms with peak area >1 � 107 km2 and duration onthe order of several weeks. In comparison, large dust storms originating in the southern hemisphere aretypically much smaller, except notably in the two cases that expanded to global scale (the 2001 and 2007global storms).

Distinct locations of preferred storm origination emerge from the dust storm image sequences, includ-ing Acidalia, Utopia, Arcadia and Hellas. A route (trajectory) ‘graph’ for the observed sequences is pro-vided. The routes are highly asymmetric between the two hemispheres. In the south, for non-globaldust storms, the main routes are primarily oriented eastwest, whereas in the north, the routes are pri-marily north–south and zonally-concentrated into meridional channels. In a few impressive cases, stormsoriginating in the northern hemisphere are observed to ‘‘flush’’ through Acidalia and Utopia, across theequator, and then branch in the low- and mid-southern latitudes. The origin of the 2007 global dust stormis ambiguous from the imaging data. Immediately prior to the global storm, a dust storm sequence fromChryse is identified. This storm’s connection to the explosive expansion observed to start from Noachis/West Hellas is unclear due to image coverage. This paper further identifies and describes three differentstyles of dust storm development, which we refer to as ‘‘consecutive dust storms’’, ‘‘sequential activation’’and ‘‘merging.’’ The evolution of a given dust storm sequence can exhibit different combinations of thesegrowth styles at different stages of development. Dust storm sequences can overlap in time, which makesthem good candidate to grow into larger scale.

� 2013 Elsevier Inc. All rights reserved.

1. Introduction

Dust strongly interacts with the martian atmospheric circula-tion. It absorbs solar radiation and absorbs/emits in the infrared,affecting atmospheric thermal structure and dynamics. The rela-tive importance and precise nature of dust lifting mechanisms re-main uncertain, but wind stresses associated with motions on

scales from regional to microscale (convective) are likely crucial(Kahn et al., 1992; Newman et al., 2002). The observed dust activityin the martian atmosphere span a wide size range from dust devilsto local, regional and planet-encircling dust storms. Only the larg-est of these storms are observable from the Earth (Martin and Zur-ek, 1993). Martin and Zurek (1993) classified dust storms with longaxis >2000 km as non-local dust storms which included regionaland planet-encircling dust storms. Cantor et al. (2001) defined re-gional dust storms in Mars Global Surveyor (MGS) Mars ObserverCamera (MOC) images as those with areas greater than1.6 � 106 km2 and lifetimes of more than 3 martian days (hence

H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127 113

forth referred to as ‘‘sols’’). Episodic planet-encircling dust stormscover contiguous latitudinal bands and can sometimes developinto global scale (the definition of a ‘‘global dust storm’’ becomesa somewhat arbitrary semantic issue). Large-scale (large regionaland global) dust storms last for weeks to months and have signif-icant impact on global atmospheric structure and circulation (Mar-tin and Richardson, 1993; Smith et al., 2002; Wang et al., 2003;Cantor, 2007; Strausberg et al., 2005; Wang, 2007). Previous stud-ies show that a large-scale dust storm results from the aggregationof many smaller dust storms or (equivalently, depending on thedefinition of the word ‘‘storm’’) from the activation of many differ-ent dust-lifting centers (Martin and Zurek, 1993; Cantor et al.,2001; Strausberg et al., 2005; Cantor, 2007; Wang, 2007; Hinsonand Wang, 2010; Hinson et al., 2012).

While prior studies have been able to identify the quasi-period-icity of global storms (Martin and Zurek, 1993), the telescopic re-cord is not of sufficient spatial resolution to determine the‘‘climatology’’ of their smaller precursor events. Developing anunderstanding of the ‘‘life-cycle’’ of these precursor storms isimportant for a few reasons: (1) While the limited observationsto date suggest that large-scale dust storms develop from smallerones, we do not yet have an empirical or descriptive understandingof why some storms grow and some do not. (2) The field of atmo-spheric modeling is now at a stage where a variety of dust stormtypes can be simulated (i.e. emerge) in numerical models (New-man et al., 2002; Toigo et al., 2002; Basu et al., 2006; Hollingsworthand Kahre, 2010; Mulholland et al., 2013). (3) The global observa-tions of Mars from MGS and Mars Reconnaissance Orbiter (MRO)have been ongoing with only brief breaks since the early northernsummer of Mars year 24 (1999). As such, with almost 7 martianyears of global imaging completed, we are now at a point wherea first attempt at a climatological survey of large-scale dust stormorigin and evolution is statistically worthwhile.

In this study, we address the following important questions:

� Are there statistical groupings of dust storm sequences thatform large dust storm families?� Are there differences in the seasonality of the families?� Where do the large dust storm families originate?� How do large dust storms evolve and do they tend to move from

specific source regions and if so, to where?

Strong constraints on the range of all possible progenitors ofglobal storms remains beyond the scope of current data sets asonly two global events have been captured so far.

In this paper, we use the Mars Daily Global Map (MDGM)products derived from MGS MOC (Wang and Ingersoll, 2002)and MRO Mars Color Imager (MARCI) wide-angle camera globalmap swaths. The period examined covers portions of Mars years24–30 (1999–2011). In order to study large-scale dust stormevolution, we focus on storms that last for at least 5 sols and in-volve dust transport and/or lifting in regions beyond their origi-nation areas. We will call the resulting selected data sets ‘‘duststorm sequences,’’ with an individual dust storm sequence ofMDGM images corresponding to each individual large-scale duststorm identified in Mars years 24–30 period. Examples will beshown in Section 5. The vast majority of the storms we studyhave peak areas >1.6 � 106 km2, satisfying the Cantor et al.(2001) criteria for a regional dust storm, although a few areslightly smaller (but still on the order of 106 km2). Since theidentified sequences cover areas beyond their origination region,they have the potential to grow larger. Two of the 65 cases iden-tified grew into global dust storms.

Each dust storm sequence we examined involved multiple lift-ing sites with various sizes and lifetimes. The association of theselifting centers with an individual large dust storm is thus retro-

spective in the analysis. By our definition, the evolution of a duststorm sequence involved one or more active lifting centers on eachday, generated a cumulative dusty area involving several con-nected regions, and had a continuous development history and awell-defined route for travel. Within a sequence, a subsequent dustlifting event could initiate either in the original initiation region oralong the route of an earlier event, and subsequent dust cloudswithin a storm sequence could then either continue along the samedirection as its predecessor or turn in a different direction. Thismultiple, parallel and dynamic character of large dust storms notonly presents a challenge for individual storm categorization, butappears to be sufficiently ubiquitous a trait of large dust stormsthat it would need to be reproduced by any successful model. Itshould be noted that more than one sequence can overlap in time(see Section 5). Consequently, there are cases when some duststorms contribute to one sequence while others contribute to an-other on the same day. When metrics such as the area for each se-quence is calculated, it is done separately on the basis of ourjudgment of dust storms as being part of the same sequence or not.

Dust storms occur frequently in the circumpolar latitudes andmany qualify as regional dust storms (Cantor et al., 2001; Cantor,2007; Wang, 2007; Wang et al., 2011). If we were to define a ‘‘duststorm sequence’’ purely in terms of dustiness over an extendedperiod of time, it would be difficult to decide whether to groupthe observed circumpolar events into one season-long ‘‘duststorm’’ or many shorter dust storm sequences. A further complica-tion of the analysis of polar storms is that the winter polar nightobviously cannot be imaged in the visible wavelengths. Therefore,we have excluded circumpolar dust storms from this study unlessthey show substantial equator-ward motion and growth outside ofthe polar region. This is not to diminish the importance of circum-polar storms as it should be noted that these storms are amongstthe most frequently observed forms of dust activity on Mars (Can-tor et al., 2001; Cantor, 2007; Liu et al., 2003) and ‘‘seed’’ the devel-opment of many larger storms we observe outside of the polarregion (and which we then do include in our study when they growand migrate in such cases).

For the limiting extreme of global dust storms, all dust liftingactivity at all locations may be considered to comprise a singledust storm sequence based on our definition. However, as de-scribed in Strausberg et al. (2005) and Cantor (2007), there weremany distinct dust lifting centers that persisted for weeks duringdifferent stages of the 2001 global dust storm and it is not clearhow these centers related to one another. In particular, it is notclear how the presence or absence of a global storm would mod-ify the ‘‘normal’’ development of large-scale dust storm (such asthat seeded by the northern frontal storms). Given that dramaticchanges in the global mean atmospheric opacity likely signifi-cantly modify the background circulation of the atmosphere,we make no attempt to interpret or include separate dust stormsequences during either of the two global dust storms. However,it should be noted that such sequences could be identified.

It should also be noted that in this study we do not includeshort-lived storms or events that remain localized – this exclusionstems directly from our definition of the problem: we seek to ad-dress how to large-scale dust storms grow, evolve, and migrate.Common areas of isolated and/or short-lived storms include theHellas basin and the Arcadia/Amazonis region to the northwestof Tharsis. In assessing all the MDGMs, we note that these seem-ingly random and isolated short-lived events can occur at manylocations. The majority should probably be considered as localevents, as cataloged by Cantor et al. (2001) and Cantor (2007), eventhrough some may have exceeded prior area-based definition of lo-cal storms. However, in our selection of events we believe that wehave included all candidate events from Mars years 24–30 thathave the potential to grow into major dust storms.

114 H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127

In this paper, we initially describe the MDGM product, extend-ing the prior application of global mosaic creation from MGS MOCwide-angle images to the MRO MARCI dataset. In Section 3, we de-scribe the seasonal evolution of dust storm sequences, includingthe seasonality of storm area, northern versus southern hemi-sphere storm origin, and the nature of dust lifting events (espe-cially the importance of frontal development). In Section 4, wediscuss the evolutionary pathways or geographical ‘‘routing’’ ofthe storms observed in the image sequences. In Section 5, we de-scribe the distinct families of storm development styles thatemerge, the occurrences of sequence overlap and their relationshipto large-scale dust storms, and the origin and evolution of the 2007global dust storm. Finally, in Section 6 we provide a summary ofresults.

2. Data sets

The basic dataset used in this study is the Mars Daily GlobalMap archive that now extends from Ls � 150� in Mars year 24 toLs = 360� in Mars year 30. The archive includes processed imagesfrom two separate orbiter camera systems – the entire MGS MOCset and the MRO MARCI set collected until late 2011. Each mapin the MDGM archive is a mosaic of up to 13 separate global imageswaths taken on a given sol by MOC (Mars years 24–28, at 2 PM) orMARCI (Mars years 28–30, at 3 PM). Consecutive MDGMs usuallyoverlap by one orbit. A MDGM is thus not a snapshot, but repre-sents a daily view of Mars, in the early-to-mid afternoon local time,gradually built up by the cameras. While this method can miss ra-pid processes (on timescales less than a sol), the dust storm se-quences that are the focus of this study are well observed inMDGMs due to their large extents and long lifetimes.

The procedure for making MDGM mosaics from the MGS MOCwide-angle images was described in detail in Wang and Ingersoll(2002). The image processing for MDGM creation from MRO MAR-CI wide-angle images follows approximately the same steps. MAR-CI is a ‘‘push frame’’ wide-angle camera with 2 UV (260 and320 nm) and 5 visible filters (425, 550, 600, 650 and 725 nm), tak-ing 1–10 km/pixel images at about 3 PM local time during each or-bit (Bell et al., 2009). We first use the USGS Integrated Software forImages and Spectrometers (ISIS, isis.astrogeology.usgs.gov) to per-form radiometric calibration and calculate the geometry and posi-tion of each pixel. ISIS currently uses the gain and offset state listedin the image label, which corresponds to the value at the beginningof the image. However, the gain sometimes changes during an im-age, leading to some uncorrected gain jumps. We have recentlycoded our own radiometric calibration program in IDL based onthe description in the ‘‘marcical.txt’’ file that accompanies eachMARCI volume downloaded from the Planetary Data System(PDS, psd-imaging.jpl.nasa.gov). Our program automatically takesinto account the possible gain changes listed in the ‘‘varexp.tab’’ ta-ble. Our new code also calculates the incidence angle, emission an-gle, phase angle, latitude and longitude of each pixel usingfunctions from the Navigation and Ancillary Information Facility(NAIF) ICY toolkit and SPICE kernels (naïf.pds.nasa.gov). This IDL-based code has begun to be used for processing the images towardthe end of the third MARCI mapping year (Mars year 30).

In the next step, we correct the vertical streaks due to pixel-to-pixel variation in each image using a normalized correction frame.Each line of the correction frame is derived from the average of thecorresponding lines in the middle third of the image. Then, we useEq. (1) with the parameter values in Table 1 to photometrically‘‘flatten’’ the image to minimize large-scale brightness variations.This equation is modified from the one used for MGS MOC imageprocessing in Wang and Ingersoll (2002), which is, in turn, similarto Hapke’s bi-directional surface reflectance function (Hapke,

1986). The parameter values were derived using the average ofabout 100 randomly selected MARCI images in mission sub-phaseB06 (February 2009). B06 was used as our reference since it wasthe first batch of MARCI images we processed. As a result, the pho-tometric correction works better for some images than others. Newparameter values may need to be derived for other time periods.For simplicity, however, we have applied a single set of parametervalues (Table 1) throughout the processing.

rði; e; g; xÞ ¼ l0

lþ l0� Hðc;l0Þ � Hðc;lÞ � ½f � Gðk1; gÞ þ ð1� f Þ

� Gðk2; gÞ� � 1þ B0

1þ h1 � tanðg=2Þ

� �� ða0 þ a1x

þ a2x2 þ a3x3 þ a4x4Þ ð1Þ

where i is incidence angle, e is emission angle, g is phase angle, x isthe pixel number along each line of the image, l0 = cos(i), l = cos(e),H(c, l) = (1 + 2l)/(1 + 2cl), G(k, g) = (1 � k2)/(1 + k2 + 2k cos(g))3/2 isthe Henyey–Greenstein phase function. Parameters used in this pa-per are listed in Table 1. They should be updated as needed. Forphotometric correction, each radiometrically calibrated pixel is di-vided by the value of r(i, e, g, x) calculated using Eq. (1).

For each color band, we mosaic 13 (or the maximum number ofswaths available for a given sol if some are missing) images foreach filter into a daily global map (90�S–90�N, 180�W–180�E,0.1� � 0.1�). The pixels in the overlap areas are created from theweighted averaging of the pixels of the contributing images. Largerweights are prescribed for the pixels with a smaller sum of inci-dence, emission and phase angles in order to blend in. ConsecutiveMDGMs usually overlap by one image swath. Finally, we combine3 MARCI visible bands (usually Band 1, 3 and 5 – blue, orange, andNIR) to form a color MDGM. An example MARCI MDGM is shown inFig. 1.

We have visually inspected the currently available MOC andMARCI MDGMs to identify and catalog dust storm sequences. Wehave focused on the area between 60�S and 60�N. Our catalog con-tains information on the solar longitude (Ls) at the beginning andend of each sequence, the surface area apparently covered by thickdust (that completely obscures the surface) on each sol during thesequence, whether the sequence originates in the southern ornorthern hemisphere, and whether it involves any curvilinearshaped ‘‘frontal’’ structure (Wang et al., 2005). We have studiedthe spatial, temporal and size distributions of these dust storm se-quences. We have also summarized their routes and developmentstyles.

To provide auxiliary information, we also use selected temper-ature and/or optical depth data derived from the MGS ThermalEmission Spectrometer (TES, Smith, 2004), Mars Odyssey THermalEMission Imaging System (THEMIS, Smith, 2009) and MRO MarsClimate Sounder (MCS, Kleinboehl et al., 2009). The version 2 TESand version 2 MCS derived data products were downloaded fromthe Planetary Data System website (atmos.pds.nasa.gov) Mars ar-chive page. The gridded THEMIS data were provided by MichaelD. Smith (personal communication).

3. Seasonality

3.1. Temporal distribution

Fig. 2 shows the Ls of occurrence and duration of dust storm se-quences identified in the MDGMs for Mars years 24–30. Missingdata periods for each year are indicated by blanks. Each sequenceis indicated by a left bracket, with the vertical line correspondingto the Ls at the beginning of the sequence and the other two slopedlines corresponding to the duration of the sequence. Each sequenceis also indicated by a filled bar.

Table 1Parameters used in Eq. (1) for photometric correction of MARCI visible images.

c k1 k2 f B0 h1 a0 a1 a2 a3 a4

Band 1 1.5 0.368 0.0 0.6907 1.248 8.9 0.2036 0. 0. 0. 0.Band 2 0.725 0.808 �0.0145 0.833 0.879 4.695 0.466 0. 0. 0. 0.Band 3 0.55 0.757 �0.001 0.834 1.718 1.865 0.306 0. 0. 0. 0.Band 4 0.40 0.822 �0.0217 0.826 0.744 2.426 0.476 0. 0. 0. 0.Band 5 0.33 0.76 �0.0216 0.75 0.611 5.0 0.36 0. 0. 0. 0.

Fig. 1. Example MRO MARCI Mars Daily Global Map (Ls � 239.5� in Mars year 30) insimple cylindrical projection (90�S–90�N, 0.1� � 0.1�).

H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127 115

This figure shows the generally excellent coverage provided bythe two camera systems over the observational period. It also pro-vides a good visual sense of the relative occurrence of dust stormsoriginating in the northern (black) and southern (green) hemi-spheres. Our designation of a given event being a ‘‘northern hemi-sphere’’ or ‘‘southern hemisphere’’ sequence is based solely on theorigination of the first dust storm of the sequence (which we inter-pret as being the ‘‘seed’’ or origination event); the designation doesnot provide information on where the dust storm or other liftingcenters are located later in the sequence. Thus a dust storm origi-nating in the northern hemisphere that is observed to lead to sub-sequent dust lifting in or transport to the southern hemisphere(e.g. Wang et al., 2003; Cantor, 2007) is still designated as a ‘‘north-ern hemisphere’’-originating event.

Fig. 2. Seasonal distribution of dust storm sequences observed in MGS MOC (Mars yearsedge dust storms at high latitudes are neglected in this study unless they move to lower lwith data but without any dust storm sequences are indicated by orange bars. Each identhe beginning and the other two sloped lines corresponding to the duration of the sequento its origination hemisphere. Northern hemisphere originated sequences are in blacksequence can travel to the southern hemisphere and vice versa. (For interpretation of thethis article.)

The global dust storms that started at Ls � 180� in Mars year 25(2001) and at Ls � 260� in Mars year 28 (2007) correspond to thelong green bars and brackets in Fig. 2. They cover the initiation,peak and initial decay phases of the global dust storms. It shouldbe noted that dust haze continued to decay after the periods indi-cated in Fig. 2. So, the lengths of the bars and brackets do not coverthe whole lifetimes of the global dust storms. Since the analysis re-lies on an operators’ subjective judgment on the border and thick-ness of a dust storm, it is difficult to tell when a global haze hascleared to the normal background condition. We therefore do notprovide an end Ls for a global dust storm here. The 2001 global duststorm originated from west Hellas (Smith et al., 2002; Strausberget al., 2005; Cantor, 2007). As will be shown in Section 5, the2007 global dust storm was immediately preceded by a dust stormtraveling from Chryse to Margaritifer Terra and probably reachingNoachis Terra. It is unclear how this Chryse storm is related to the2007 global storm, for which continuous lifting began betweenHellas and Noachis (Section 5). As such, the 2007 global dust stormis colored green in Fig. 2, indicating a southern-hemisphere origin.

Fig. 2 shows that dust storm sequences are most frequently ob-served during Ls � 130–250� and Ls � 305–350�. During Ls � 250–300�, there is only a single case: the 2007 global dust storm in Marsyear 28. No dust storm sequences are identified in the MDGMs dur-ing Ls � 40–130�. During Ls � 0–40�, there are only two cases, oneoriginating in Acidalia in the northern hemisphere in Mars year 26and the other in Solis/Bosporus in the southern hemisphere inMars year 28. In general agreement with previous results (Martin,1986; Martin and Zurek, 1993; Liu et al., 2003), Fig. 2 suggests thatthe northern spring and summer are relatively free of large dust

24–28) and MRO MARCI (Mars years 28–30) Mars Daily Global Maps (MDGM). Capatitudes. Periods with missing MDGMs for each year are indicated by blanks. Periodstified sequence is indicated by a left bracket with the vertical line corresponding toce. Each dust storm sequence is also indicated by a filled bar color-coded according

and southern hemisphere originated sequences are in green. Note that a northernreferences to color in this figure legend, the reader is referred to the web version o

f

116 H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127

storms, and the opposite seasons are characterized by sporadiclarge dust storms that vary from year to year. In their review ofdust activity before the 1990s, Martin and Zurek (1993) found thatlarge storms (storms larger than local events) only occurred duringLs � 160–330�. Our results show a somewhat wider seasonal win-dow for non-local storms that adds the adjacent periods fromLs � 130–160� and Ls 330–350�. This was also noticed in Cantor(2007). However, as will be shown later (Section 5), all of the larg-est dust storm sequences (with peak area >1 � 107 km2) are ob-served during Ls � 130–320�.

Fig. 2 shows that the northern sequences (black) are mainlyconcentrated in two seasonal windows – the northern fall(Ls � 180–250�) and the northern winter (Ls � 305–350�). Thereare typically several instances each year within each window. Aswill be shown later, the majority of these dust storms are relatedto frontal dust storms that later develop into flushing events(Wang et al., 2003, 2005; Wang, 2007). The northern sequence atLs � 143–148� in Mars year 29 was composed of daily dust stormin Chryse and showed limited eastward motion toward the Arabiaborder.

In addition to the two global dust storms, Fig. 2 shows that thesouthern sequences (green) are predominantly observed from midto late southern winter (Ls � 130–180�). There were two non-glo-bal storm southern hemisphere sequences observed in mid-south-ern spring. One started from Hellas at Ls � 224� in Mars year 27,and later merged with a sequence from Chryse (Section 5). Theother also started from Hellas and traveled to Noachis duringLs � 220–224� in Mars year 28.

Besides the difference in seasonality between the northern andsouthern sequences, there is also a difference in duration. Fig. 2indicates that most southern sequences tend to be well separatedin time and last for either only a few days (5 or more, by our def-inition) or several months (global dust storms). In comparison, thenorthern sequences sometimes form clusters and last either a fewdays or a few weeks. This will be discussed further in the next sec-tion. Although none of the northern sequences are observed to de-velop into global dust storm, some of them can develop into

Fig. 3. Daily total areas of thick dust contributing to dust storm sequences as a functionsequence is connected with a line. In the top panel, northern sequences are colored blainvolve frontal dust storms are colored red, sequences that do not are colored black andthis figure legend, the reader is referred to the web version of this article.)

planet-encircling scale and lead to pronounced perturbations ofthe atmospheric structure and circulation (Wang et al., 2003; Liuet al., 2003; Smith, 2008).

Fig. 2 also shows some inter-annual variability of dust storm se-quences. Global dust storms occurred in Mars years 25 and 28 atLs � 180� and 260�, respectively. The onset of the northern se-quences following the global dust storms in Mars year 28 appearedto be delayed. Of the non-global dust storm years, Mars year 24exhibited the longest hiatus between the northern fall stormsand northern winter storms. In the northern fall of Mars years 28and 30, there was a lack of weeks-long dust storm sequences.

3.2. Ls–area distribution

Fig. 3 shows the Ls versus area distribution of the dust storm se-quences identified for all the Mars years examined. Again, a se-quence is defined as lasting for P5 days and influencing areasbeyond their origin. On each day, there may be one or more dis-crete dust hazes or storms contributing to a sequence. We manu-ally mark these dusty areas by judging the albedo contrast withthe surroundings, the morphologies, the day-to-day variationsand whether or not dust is thick enough to block the view of sur-face features. Each data point in Fig. 3 represents the total area ofthe contributing thick dust on that day. The data points for each se-quence are connected. The area shown in Fig. 3 should be consid-ered as a lower limit since it only includes the part of dust thatappear to severely obscure the surface in MDGMs. The whole life-cycle of each sequence is plotted except for the global dust stormswhose areas run off the top of the scale at Ls � 180� and 260�,respectively. Of the 65 total dust storm sequences, 62 have peakareas >1.6 � 106 km2. The other 3 are only slightly below this limit,but their areas are still on the order of 106 km2. Therefore, the vastmajority of the dust storm sequences we study satisfy the defini-tion of regional dust storm in Cantor (2007). All the sequences withpeak areas >5 � 106 cm2 are observed during Ls � 130–350�, andthose >1 � 107 cm2 are observed during Ls � 130–320�.

of Ls. All the sequences identified in MDGMs for Mars years 24–30 are plotted. Eachck and southern sequences are colored green. In the bottom panel, sequences thatambiguous cases are colored orange. (For interpretation of the references to color in

Fig. 4. The peak area versus duration of dust storm sequences. The 2001 and 2007global dust storms are not included as they would be off the scale. The top panelshows the result of a cluster analysis and the regression line for each group. In thebottom panel, northern sequences are colored black and southern sequences arecolored green. (For interpretation of the references to color in this figure legend, thereader is referred to the web version of this article.)

H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127 117

In the top panel of Fig. 3, the northern sequences are coloredblack, and the southern sequences are colored green. Except forthe global dust storms that explosively expand from the southernhemisphere, the northern sequences can be substantially largerthan the southern sequences. Both the northern and southern se-quences show preferred seasonal windows, as discussed in Sec-tion 3.1. Though some sequences stay within their originationhemispheres, others cross the equator at some points. In fact, largenorthern hemisphere-originating sequences often attain theirmaximum dust loading and extent in the southern hemisphere.

In the bottom panel of Fig. 3, the sequences associated withfrontal dust storms are colored red, those exhibiting no frontal fea-tures are colored black, and ambiguous cases are colored orange.As long as there is at least one clearly identifiable frontal duststorm (with characteristic curvilinear shape) contributing to a se-quence, regardless of where and when, then the sequence is col-ored red in this figure. Obviously, such a sequence can still haveother types of dust lifting/storm morphology during its historyand those other morphologies may or may not occur more oftenthan frontal dust storms. By comparing the top and bottom panel,it can be inferred that most of the northern sequences are associ-ated with frontal dust storms. The few larger ones are associatedwith flushing events (Cantor et al., 2001; Cantor, 2007; Wang,2007; Hinson and Wang, 2010). These fronts are generally associ-ated with the northern polar vortex and are believed to result frombaroclinic instability, analogous to low-pressure frontal cycloneson the Earth (Wang et al., 2003). Curvilinear fronts could also de-velop along the cap edge due to the ‘collision’ of air masses offthe cap with the surroundings (Toigo et al., 2002). A number ofsouthern hemisphere sequences are also associated with frontaldust storms. Although within a given dust storm sequence, notall the component dust storms are frontal in nature (or remainfrontal for the entire sequence), Fig. 3 suggests that frontal devel-opment is an important link in the evolution of most dust storm se-quences. However, frontal dust storms are not involved in all duststorm sequences. The black lines in the bottom panel of Fig. 3 showa significant minority that exhibits no frontal/curvilinearstructures.

Fig. 3 suggests that the sequences with larger peak areas aregenerally also the ones that last longer. However, some sequencesthat attain only moderate peak areas can also persist for extendedtime periods. This is better shown in the scatter plot of peak areaversus duration measured in degrees of Ls (Fig. 4).

In the top panel of Fig. 4, a cluster analysis based on the peakarea and duration divides the sequences into the following threegroups. Note that other choices of clustering criteria and numberof output groups will lead to different results.

� Group 1 contains the two global dust storms (which are off thescale of the plot, therefore not shown);� Group 2 (red) contains 9 sequences that cluster around the

regression line of y = 7.5 � 105x + 1.2 � 107 and all the membershave peak area >1 � 107 km2; Since the number of data pointsin this group is limited, the regression line is likely not robust,but the existing data suggest a positive correlation betweenthe area and duration for the sequences in this group.� Group 3 (blue) contains the rest, including the sequences with

relatively short durations and small areas, as well as the 5sequences that are >8� of Ls but <1 � 107 km2, which are in con-trast to the members of Group 2.

In the bottom panel of Fig. 4, the northern sequences are col-ored black and the southern sequences are colored green. This pa-nel shows that southern sequences that are not global storms aregenerally short-lived, with durations of less than 8� of Ls. Therethus appears to be a lack of southern population with intermediate

duration (southern storms are either relatively short lived or largeglobal events). It can also be seen that the three largest non-globalsouthern sequences with a peak area exceeding 1 � 107 km2 lastless than 6� of Ls. These sequences were observed in MDGMs dur-ing Ls � 136–138� in Mars year 27, Ls � 224–230� in Mars year 27and Ls � 152–157� in Mars year 29.

In comparison, while northern sequences have not been ob-served to grow to global extent, the northern population includessequences with intermediate as well as shorter durations. Amongthe northern sequences with intermediate durations of 7–15� ofLs (2 weeks–1 month), one set has peak area >1 � 107 km2. It iscomposed of sequences that involve cross-equatorial dust stormsand that develop toward or into planet-encircling scale. They in-clude one storm that traveled through Acidalia during Ls � 220–230� in Mars year 24 (Cantor et al., 2001; Wang et al., 2003), onethat traveled through Utopia during Ls � 205–219� in Mars year26, one through Acidalia during Ls � 313–322� in Mars year 26(Cantor, 2007; Wang, 2007), one through Acidalia duringLs � 214–228� in Mars year 27, one through Acidalia duringLs � 313–322� in Mars year 27 (Hinson and Wang, 2010) and onethrough Acidalia during Ls � 230–237� in Mars year 29.

The remainder of the northern hemisphere, intermediate life-time dust storm sequences have peak areas <1 � 107 km2. In somecases, dust storms were repeatedly observed for many days along arelatively short route, for example, in the Acidalia–Chryse channel.In other cases, dust storms progressed further each day following a

Fig. 5. Histogram of the origination regions of dust storm sequences observed in MGS and MRO MDGMs during Mars years 24–30. These regions are indicated in the MDGMsin Fig. 7.

118 H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127

long route, but only part of the long route was covered by thickdust on a particular day. However, in both types of cases, duststorms within the sequences were not able to initiate furthergrowth. As a result, the dust storm sequences stayed much smallerthan planet-encircling scale. The various development styles ofdust storm sequences will be discussed in Section 5.

4. Origin locations and storm travel routes

4.1. Origination areas

Fig. 5 shows the histogram of the origination areas of the duststorm sequences. The northern sequences are colored black, andthe southern sequences are colored green. Acidalia is the regionwhere the most dust storm sequences originate, followed by Uto-pia, Arcadia and Hellas. Other regions include Solis/Bosporus, Cim-meria/Sirenum, Chryse, Argyre, Noachis, Margaritifier and thesouthern edge of Arabia. Most of these areas have previously beenidentified as frequent dust storm source regions (Martin and Zurek,1993; Cantor et al., 2001; Cantor, 2007). The number of sequencesfrom Acidalia is about three times of that from Utopia. Hinson et al.(2012) also found that Acidalia had much more frontal dust stormactivity than other areas. Hinson and Wang (2010) suggested thefollowing factors that affect the timing and location of regionalfrontal/flushing dust storms: (1) transitions among baroclinic wavemodes, (2) storm zones and (3) stationary waves. While Acidalia isthe most preferred region for the northern sequences, Hellas is themost preferred site for the southern sequences. In agreement withthese observations, Mars GCM simulations suggest that the area

Fig. 6. Routes for dust storm sequences observed in MGS and MRO MDGMs during Marsgreen. Dashed lines at the top and bottom indicate that circumpolar dust storms are nstorms. The number of sequences through each segment is indicated in the plot. (For inteweb version of this article.)

north of Tharsis and upwind of Acidalia and the area in westernand northern Hellas have strong surface stresses (Newman et al.,2002; Basu et al., 2006; Kahre et al., 2006; Toigo et al., 2012; Mul-holland et al., 2013).

4.2. Routes

In Fig. 6, the observed routes traveled by dust storm sequencesare plotted on a storm-free MDGM (60�S–60�N). This figure sum-marizes the variety of cases observed during Mars years 24–30and indicates where dust storm sequences originate and wherethey go. The diagram is in the form of a graph (e.g. Hartsfeld andRingel, 2003). The lines and arrows form unidirectional edges indi-cating the general routes. Note that the edges do not show the ex-act pathways of any specific sequence. Northern sequences arecolored black and southern sequences are colored green. Thedashed lines at the top and bottom indicate that circumpolar duststorms are neglected in this study. The dashed arrows near Hellasindicate the initial expansions of the global dust storms in Marsyears 25 and 28. Subsequent development of global dust stormsis not considered in this paper. The numbers in the graph indicatethe number of sequences passing along the corresponding edge. Adust storm sequence can start from any edge, follow one or moresubsequent edges, and terminate anywhere along the main routeor branches. Consequently, the sum of numbers for the branchesis different than the number for the main route.

Fig. 6 shows that the routes are distributed in areas where GCMsimulations predict high surface wind stresses (Newman et al.,2002; Basu et al., 2006; Kahre et al., 2006; Toigo et al., 2012;

years 24–30. Northern sequences are colored black. Southern sequences are coloredeglected in this study. Dashed arrows indicate the initial expansion of global dustrpretation of the references to color in this figure legend, the reader is referred to the

H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127 119

Mulholland et al., 2013). The main east–west routes are in thesouthern hemisphere and the main north–south routes extendfrom the northern to the southern hemisphere. Although thereare north–south routes from the southern hemisphere as well, theyare generally shorter (except for those active during the global duststorms) and occur less frequently. Since planet-encircling duststorms imply longitudinal expansion to cover a latitudinal band,they are most likely to go through the southern hemisphere toachieve this status. However, many east–west routes are branchesof the north–south routes and some north–south routes are ob-served to stretch over considerable latitudinal distances. Conse-quently, some nearly planet-encircling dust storms can be tracedback to the northern hemisphere.

4.2.1. Northern sequencesFig. 6 shows that the three main routes from the northern to the

southern hemisphere are through Acidalia, Utopia and Arcadia.These sequences involve multiple frontal/flushing dust storms, asdescribed in previous studies (Wang et al., 2005; Hinson andWang, 2010). The majority of them dissipate in the northern hemi-sphere. However, a few impressive sequences go across the equa-tor to the southern hemisphere where they dissipate or followbranches.

For the Acidalia sequences, one branch turns eastward in thesouthern low latitudes toward the corridor between Arabia andNoachis. An example can be found during Ls � 220–230� in Marsyear 24 (Cantor et al., 2001; Wang et al., 2003). The other branchcontinues further south toward Argyre and beyond, turns towardHellas and/or Solis. An example can be found during Ls � 313–322� in Mars year 26 (Cantor, 2007; Wang, 2007).

A few Utopia sequences continued across the topographicdichotomy. One sequence (Ls � 233–236� in Mars year 28) turnedslightly eastward toward Hesperia/Cimmeria. Two sequences(Ls � 206–219� in Mars year 26 and Ls � 305–314� in Mars year29) turned westward and went around the area north of Hellas to-ward Noachis and beyond. Both routes are consistent with the sur-face wind directions simulated by GCM for the correspondingseasons (Toigo et al., 2012).

For the Acidalia and Utopia sequences that travel very long dis-tances from the north to the south, while the early stage dust lift-ing activity is clearly in the northern hemisphere, later on in thestorm evolution, the major dust loading and lifting centers shiftto the southern hemisphere. The ones that exhibit longitudinalexpansion in the southern hemisphere can grow into planet-encir-cling dust storms (Cantor et al., 2001; Cantor, 2007; Wang et al.,2003; Wang, 2007). The relationship of a briefly observed Acida-lia/Chryse storm to the origin of the 2007 global storm is highlyambiguous (see Section 5.3).

The longest Arcadia sequence observed so far was duringLs � 226–231� in Mars year 24 (Fig. 8). It traveled southward acrossthe equator to Sirenum/Cimmeria. Although the flushing duststorms in the Arcadia sequences appeared as impressive as thosein Acidalia or Utopia sequences (Wang et al., 2005), they are ob-served less frequently. Within our data set, the Arcadia sequenceusually covers shorter latitudinal range and exhibits no or littlelongitudinal expansion.

4.2.2. Southern sequencesHellas is the most important origination area of the southern se-

quences observed so far. Routes go in many directions toward Hes-peria, Elysium, Syrtis, Noachis and the south polar cap. The 2001global dust storm originated from the western and northern slopesof Hellas. It exhibited rapid eastward expansion to Hesperia andnorthward expansion to Syrtis at the beginning, and later involveda second major dust lifting center in the Syria/Solis area later on(Smith et al., 2002; Strausberg et al., 2005; Cantor, 2007). For the

2007 global dust storm, the first major dust lifting was in the Noa-chis/West Hellas area, though there was a preceding (and ambigu-ously related) sequence from Chryse (Section 5).

The region enclosed by Syria, Argyre and Valles Marineris (la-beled as Solis/Bosporus in Fig. 5) is also a common origination areafor southern sequences, so is the Cimmeria/Sirenum region. Thecorresponding sequences generally move northward toward theequator. Some Solis/Bosporus sequences appear to travel alongValles Marineris. Martin and Zurek (1993) pointed out that the Syr-ia/Solis area was an important dust source region for global duststorms, but not for local dust storms. Consistent with their results,our observations show that the Syria/Solis area seldom had localdust storms, but was active after the initiation of both 2001 and2007 global dust storms and was also involved in several otherdust storm sequences. This behavior suggests that dust lifting inthe area may require enhancement of the circulation resultingfrom the collective effect of other dust lifting centers in order tobecome activated.

5. Development styles

5.1. Classification

The most common development style observed for dust stormsequences is the daily repetition of dust storms traveling alongthe same route, keeping the region dusty for 5 sols or more. Wewill simply call this style of development ‘‘consecutive duststorms’’. This type can be seen in Fig. 7a–e where Days 10–14 ofa dust storm sequence during Ls � 214–228� in Mars year 27 isshown. On each day, there are one or more dust storms withinthe Acidalia–Chryse channel. This consecutive dust storm sequenceappears to be related to frequent frontal initiation (or ‘‘pumping’’of the storm by frontal systems) in the high northern latitudes.Similar frontal/flushing dust storms are observed in the area dailyfor the first 3 weeks or so of the sequence. If the generation of theflushing events is associated with coherent timing of the low-pres-sure weather systems having favorable phase of the thermal tide(Wang et al., 2003), then this extended time period probably rep-resents an extended time period of near-phase locking of thesystems.

Another development style is through sequential activation ofone segment of a route after another as the whole sequence ad-vances forward. We will call these ‘‘sequential activation duststorms’’. Example images can be found in Figs. 19 and 20 of Cantor(2007) and Figs. 2 and 3 of Wang (2007). These examples show thedevelopment of dust storm sequences originating in Utopia–Isidisduring Ls � 210–220� and in Acidalia–Chryse during Ls � 310–320� in Mars year 26. In each case, the dust activity center pro-gressed further each day along the route. As a new segment be-come dusty, dust in the previous segment dissipated. TheAcidalia sub-sequence shown in Fig. 7 can be considered to belongto this category since dust activity shifted from the northern to thesouthern hemisphere with time.

A third development style is through the merging of dust fromtwo or more initially separate sequences to create a contiguousdust cover. We will call these ‘‘merging dust storms’’. This ap-peared to be a very effective way of making larger dust storms,including the two global storms. The 2001 global storm resultedfrom the merger of dust from storms (or alternatively, lifting cen-ters) originating near Hellas and Syria-Solis area (and potentiallyalso in the northern high latitudes) (Strausberg et al., 2005; Cantor,2007). Fig. 7 shows an example for this type, too. During the courseof the Acidalia sequence, another sequence started from northernHellas (Fig. 7f) and traveled toward Noachis (Fig. 7g). In the

Fig. 7. Example MGS MDGMs (60�S–60�N, in simple cylindrical projection) for the development of the dust storm sequence during Ls � 214–228� in Mars year 27. Memberdust storms are indicated by white arrows. (a) Day 10, (b) Day 11, (c) Day 12, (d) Day 13, (e) Day 14, (f) Day 17, (g) Day 20 and (h) Day 22.

120 H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127

meanwhile, the Acidalia sequence turned toward the corridor be-tween Noachis and Arabia, and merged with the Hellas sequence.

There can be different combinations of these developmentstyles for a sequence. In another word, a dust storm sequencecan exhibit different development styles at different stages. Thisis clearly the case in the example shown in Fig. 7. The overall devel-opment is through ‘‘merging’’ of the Acidalia and Hellas sub-se-quences. The Acidalia sub-sequence contributing to the totalstorm sequence can be further classified as ‘‘sequential activation’’since the dust activity moved from the northern hemisphere to thesouthern hemisphere, but the first �2/3 of the Acidalia sub-se-quence developed through ‘‘consecutive dust storms’’ in the Acida-lia–Chryse channel. The Hellas sub-sequence also developedthrough ‘‘consecutive dust storms’’ in the Hellas–Noachis area.

5.2. Overlap

As mentioned earlier, dust storm sequences can overlap in time.For example, during Ls � 220–230� in Mars year 24, a dust stormsequence developed from Acidalia and headed toward Argyre andthe corridor between Arabia and Noachis (Cantor et al., 2001; Liuet al., 2003). During the course of this sequence, two other se-quences were observed – one from Utopia to Elysium duringLs � 221–225�, and the other from Arcadia to Cimmeria/Sirenumduring Ls � 226–231�. The Ls versus area plot of the three se-quences observed in MDGM is shown in Fig. 8a. The day-to-dayvariability shown by the curves is probably not statistically signif-icant since there are cases where dust storms gradually merge tothe background, making it difficult to decide on the border. How-ever, the relative size of the sequences is robust. The Acidalia se-quence (shown in red) had the longest duration and the largestpeak area. The Utopia sequence (shown in green) was much smal-ler and shorter. The Arcadia sequence (shown in purple) achieved

nearly the same peak area as that of the Acidalia sequence, butlasted only half as long.

Fig. 8b shows the areas affected by the Acidalia (red), Utopia(green) and Arcadia (purple) sequences as judged by visual inspec-tion of MDGMs. We identify dusty areas on each day for each se-quence. As long as a pixel is considered dusty at some pointduring the observational period of Ls � 220–230�, it is highlightedin the figure. In another word, the highlighted areas were appar-ently covered by dust on one or more days during Ls � 220–230�.Most of the areas in the northern plains were involved. All three se-quences contributed dust to the low latitudes. The Acidalia andArcadia sequences penetrated to the southern mid-latitudes.

Liu et al. (2003) tracked the dust activity described above usingoptical depth maps derived from MGS TES data for every 2� of Ls

during Ls � 205–245� in Mars year 24. Their results showed thata nearly continuous dust band formed at the southern low lati-tudes after Ls � 231�. Although TES data did not allow some ofthe details of the dust storm structures to be observed due to theirlimited spatial resolution and coverage as compared with MDGMs(Wang et al., 2013), they were able to describe the overall develop-ment history of this long lasting event.

In Fig. 8c, we have plotted the maximum TES dust optical depthfor each 4� � 5� grid box during the Ls � 220–230� period of Marsyear 24. There are typically about 300 data points in each gridbox. The low topographic channels from Acidalia, Utopia and Arca-dia show high dust amount, as well as the Corridor between Arabiaand Noachis in the southern low latitudes. The dust optical depthin the plot is not normalized to the same pressure level. However,a comparison with similar plots for dust storm free periods (notshown) clearly indicates that the high optical depths result fromthe three overlapping dust storm sequences described previously.The optical depths shown in Fig. 8c also include contributions fromsporadic dust storms and dust hazes whose effect accumulates

Fig. 8. (a) Ls versus area plot of the dust storm sequences observed during Ls � 220–230� in Mars year 24. The Acidalia sequence is shown in red, Utopia sequence in green andArcadia sequence in purple. (b) The area in MDGM (60�S–60�N) that is apparently affected by the dust in the Acidalia (red), Utopia (green) and Arcadia (purple) sequence. (c)The maximum MGS TES dust optical depth for each 4� � 5� grid box during Ls � 220–230� in Mars year 24. (For interpretation of the references to color in this figure legend,the reader is referred to the web version of this article.)

H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127 121

during the analysis window. The MDGM images indicate that theseevents are mostly near the edge of the southern polar cap andHellas, distinguishing them from the dust storm sequences.

To investigate when dust storm sequences overlap during ourobservational period, in Fig. 9 we have indicated dust storm se-quences with brackets on the Ls versus latitude map of the zonalmean atmospheric temperatures. The temperature data are fromthe 2 AM TES temperatures at 64.3 Pa (Mars years 24–26, Smith,2004), the 5 PM THEMIS temperatures at �50 Pa (Mars years

27–28, Smith, 2009) and the 3 AM MCS temperatures at 64.3 Pa(Mars years 29–30, Kleinboehl et al., 2009). Note the differencesin local time and pressure level. The vertical edge of a bracket cor-responds to the first day of a sequence. The other two edges extendto the last day of the sequence (except for global dust storms). Wehave marked the sequences larger than 107 km2 with ‘�’s on thecorresponding brackets.

Overlapped sequences are represented by overlapped bracketsin Fig. 9. It can be seen that overlap in our observational period

-50

0

50Mars Year 24TES 2AM P=64.3Pa

XX

-50

0

50Mars Year 25TES 2AM P=64.3Pa

X

-50

0

50Mars Year 26TES 2AM P=64.3Pa

X X

-50

0

50Mars Year 27THEMIS 5PM P~50Pa

X XX X

-50

0

50Mars Year 28THEMIS 5PM P~50Pa

X

-50

0

50Mars Year 29MCS 3AM P=64.3Pa

X X

0 30 60 90 120 150 180 210 240 270 300 330 360

-50

0

50Mars Year 30MCS 3AM P=64.3Pa

130.0 138.0 146.0 154.0 162.0 170.0 178.0 186.0 194.0 202.0 210.0

Fig. 9. Filled contours show the Ls – latitude distribution of the zonal mean atmospheric temperate (K) at 64.3 Pa derived from MGS TES 2 AM data (Mars years 24–26), at�50 Pa derived from Mars Odyssey THEMIS 5 PM data (Mars years 27–28) and at 64.3 Pa derived from MRO MCS 3 AM data (Mars years 29–30). Missing temperature data areleft blank. The left brackets indicate the starting Ls and duration of dust storm sequences identified in MGS MOC or MRO MARCI MDGMs. Northern sequences are coloredblack and southern sequences are colored green. Sequences with peak areas >107 km2 are indicated by ‘�’s. (For interpretation of the references to color in this figure legend,the reader is referred to the web version of this article.)

122 H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127

is mainly observed during Ls � 200–240� and Ls � 300–340�. Thereare 8 cases of overlap during these two seasonal windows, six ofthem involve sequences with peak area >107 km2, indicating thatoverlapped dust storm sequences are good candidates for expan-sion to larger scale. As mentioned in Section 3.2, we did not at-tempt to map separate dust storm sequence overlapping withthe global dust storms due to concern over the degree to which

the global events would have fundamentally changed the large-scale circulation. Had we done so, we count 10 cases of overlapduring the Ls � 175–340� window, about 80% of them with peakarea >107 km2. There is only one instance of an overlap outside thistime period. It was observed during Ls � 155–160� in Mars year 27when one sequence traveled from Cimmeria/Sirenum to Amazonisand the other from Argyre to Valles Marineris. They were relatively

H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127 123

small and located far away from one another. So, not every tempo-ral overlap necessarily led to a large dust storm. Other examplescan be found in the post-solstice period of Mars years 24 and 30.

5.3. Large sequences

The available data in Fig. 9 suggests that sequences with peakarea >107 km2 in the northern fall and winter are associated withpronounced atmospheric temperature increases. In the early stageof the 2001 global dust storm in Mars year 25, the temperature in-crease appeared to be mainly in the low latitudes. For all othercases during these two seasons, including the middle and latestages of the 2001 global dust storm, the temperature field exhib-ited a bimodal distribution with one maximum at the southernmid/high latitudes and the other at the northern mid/high lati-tudes. Similar signatures in mid-level air temperatures followingmajor dust storms had been reported in many previous studiesand attributed to dust radiative heating in the south and dynamicalheating in the north (Liu et al., 2003; Smith, 2004; Martin and Zur-ek, 1993). The MCS data show a bimodal temperature distributionin the fall of Mars year 30, though the dust storm sequences ob-served in the northern fall and winter seasons are all <107 km2.However, the peak area of the largest storm of that year (duringLs � 227–231� that year) was just short of 107 km2. Consideringthe uncertainty of the area measurements, this sequence may beconsidered to be large. The sequence during Ls � 242–249� in thatyear covered a very long latitudinal distance from Acidalia toArgyre. However, its area on each day remained relatively smallas the sequence developed in a ‘‘sequential activation’’ fashion.

There were two southern sequences with peak area >107 km2

observed in the northern spring and summer, although we mea-sured both to be just slightly above the threshold. The THEMIS 5PM atmospheric temperatures at �50 Pa showed an increase abovebackground values following the development of the dust stormsequence from Arabia to Hellas during Ls � 136–138� in Mars year27. The MCS 3 AM temperatures at 64.3 Pa showed an increase atlow latitudes as compared with those of the next year after the se-quence from Cimmeria/Sirenum to Amazonis during Ls � 152–157�in Mars year 29.

5.4. Initiation of the 2007 global dust storm

Previous image observations suggest that global dust stormsoriginate in the southern hemisphere during the southern springand summer dust storm season (Martin and Zurek, 1993; Straus-berg et al., 2005; Cantor, 2007). Two global dust storms occurredduring our observation period – the 2001 global dust storm atLs � 180� in Mars year 25 and the 2007 global dust storm atLs � 260� in Mars year 28. While the 2001 global storm was unam-biguously triggered in the mid-southern latitudes, the origin of the2007 storm was less obvious. The major growth of the 2007 stormwas at southern mid-latitudes between Hellas and Noachis. How-ever, a confined (‘‘flushing’’) dust storm was observed movingsouthwards from Chryse to Margaritifer Terra prior to the onsetof subsequent major dust lifting. This raises the possibility thatthe Chryse storm in some way triggered or merged with the south-ern mid-latitude dust storm.

Fig. 10 shows ten consecutive MARCI MDGM (90�S–90�N) nearthe beginning of the 2007 global dust storm (Ls � 261–267�). Notethat each MDGM is composed of 13 sets of consecutive global im-age swaths spaced about 2 h apart and that the 1st swath of eachMDGM is the same as the 13th swath of the previous MDGM.The figure shows that during the first 3 days of the sequence, a duststorm from Chryse traveled southward to the east of VallesMarineris. It is possible that the Chryse storm arrived in the Noa-chis region by Day 5. However, the image coverage makes it hard

to be certain, as will be explained next. The dust associated withthis flushing event appears to have dissipated after Day 5 whennew dust lifting centers near Noachis took over.

The Chryse dust storm occurred immediately prior to the onsetof the 2007 global dust storm and it is tempting to associate thisflushing event with the origin of the global storm. Ambiguity inthis association, however, arises from the nature of the data cover-age. The Chryse flushing event is captured in image Swath 13 ofDay 2 which is the same as Swath 1 of Day 3. As such, there isno new information on the event in Day 3. However, Swath 13 ofDay 3 shows a dust cloud about 2000 km to the southeast (in theNoachis region). The connection between these events is ambigu-ous. If the dust clouds were the same event caught at differenttimes (about a sol apart), and transport was the dominant pro-cesses in the continuity of the cloud (rather than new lifting), itwould require an advection speed of about 25 m/s. Such speedsare not beyond the range of possibility, but it would require theaverage wind to be at this high speed for the full day. Consideringthe wind direction rotation associated with thermal tides, muchhigher instantaneous winds would be required if they were fromthe same dust cloud (e.g., Wilson and Hamilton, 1996).

There appear to be two dust centers on Day 4, although theNoachis dust cloud is in Swath 1, which is again a repeat of Swath13 of Day 3. The other dust cloud is located to the west of the Noa-chis dust cloud. However, the seam between the swaths prevents aclean interpretation. This dust cloud could be a part of the Noachisstorm, a continuation of the flushing storm on the previous day, anew flushing dust storm or a new dust lifting center. It should benoted that on Day 5, the Noachis dust cloud remained roughly atthe same area as that on the previous day. This seems to be at oddswith a rapid advection of a flushing event to Noachis during Day 3.

The global map swaths have greatly improved our observationsof martian dust storms. However, some critical information isunfortunately lost due to the fixed local time coverage from polarorbits. Continuous monitoring from a geostationary orbit wouldeliminate the guessing, and further improve our understanding ofthe martian dust cycle.

While it is uncertain to somewhat unlikely that the dust cloudfrom the Chryse flushing event is the same as the one that appearsin Noachis on Day 3, the images cannot tell us whether a dynamicalcoupling existed between these regions. It would be interesting toexamine how the heating resulting from the Chryse storm modi-fied the circulation and whether this was in part responsible fordust lifting in Noachis that directly began the 2007 global duststorm.

It is particularly interesting to note that during the Ls � 260–270� period for Mars years 24–30, we have only identified oneChryse dust storm sequence: the event discussed here that directlypreceded the 2007 global dust storm (in Mars year 28). Indeed, itshould be noted that in the Oxford Mars GCM, flushing dust stormswere a common feature (Newman et al., 2002) and that they couldtrigger global dust storms (Mulholland et al., 2013), however, it isunclear with such small-number-statistics whether the correlationof the Chryse storm with the 2007 global dust storm is coincidenceor causation, and the putative mechanism of any linkage remainsto be elucidated beyond the speculative ‘‘enhancement’’ of thecirculation.

The remainder of the 2007 global dust storm initiation playedout directly from the Noachis dust storm. The MDGM sequenceshows the lifting in Noachis to strengthen over the next 3 sols withvery little lateral motion (Fig. 10). A threshold appears to havebeen dramatically crossed between Day 6 and 7, as the stormexplosively grew during Day 7. Multiple secondary dust lifting cen-ters appear to have become involved as early as Day 8 or 9 in theMDGM sequence. By the 10th day, the storm is well on its way to-ward planet-encircling scale. During Days 7–10, the dust rapidly

Fig. 10. MRO MARCI MDGMs near the beginning (Ls � 261–267�) of the 2007 global dust storm in Mars year 28. Each image shows the area between 90�S and 90�N in simplecylindrical projection. Images #1 to #13 indicated in each panel are sequentially taken by MARCI and pieced together in each MDGM. Consecutive MDGMs overlap by oneorbit (image #13 of the previous day is the same as image #1 of the present day). Missing data or data in error are shown in black. The sampling distribution of the MRO MCSdust optical depth at 3 AM is shown by ‘/’ (oriented northeast–southwest) and that at 3 PM is shown by ‘n’ (oriented northwest–southeast). The black arrows indicate duststorms.

124 H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127

expanded along a primarily east–west axis, but with a notablenortheast–southwest tilt, with expansion mainly to the southsouth west (SSW) and to a lesser extent to the north north east(NNE).

Fig. 11 shows the latitude versus height (km) cross-sections ofthe vertical distribution of MCS dust opacity (Kleinboehl et al.,2009) for the 10 days shown in Fig. 10. The spatial distribution ofnighttime sampling is shown by ‘/’ (oriented northeast–southwest)and that of daytime sampling is shown by ‘n’ (oriented northwest–southeast) in Fig. 10. Since the nighttime sampling has much bettercoverage, especially for the low latitudes, we have plotted the 3 AMMCS data in Fig. 11. The MARCI images were taken at 3 PM, there-fore, there is half a day offset between Figs. 10 and 11. Most of thetracks that cut across dust storms in Fig. 10 are for the nighttime,i.e., they do not coincide with the dust storms in images, but arewithin a day of their occurrences.

In Fig. 11, the 3 AM MCS dust optical depths are binned onto4� latitude � 30� longitude grid, and interpolated onto fixed

heights above local surface from the 105 fixed pressure levelsof the MCS retrieval. The zonal maxima are extracted and theirlog values plotted for the lowest 50 km. We use the zonal maxi-mum to indicate the height of dust penetration for a zonal band.Due to the sparse coverage of MCS dust optical depth at low lat-itudes (Fig. 10), the maximum is usually derived from only 2–6orbits for the day. The MCS dust retrievals typically have errorswithin 20% and are difficult under high dust opacity conditions(Kleinboehl et al., 2009). Therefore, the plotted quantity may con-tain substantial error and the details of the panels should not beover interpreted. Nevertheless, the dust top after Day 6 is consis-tently higher than that before Day 6, suggesting that the imageddust clouds are consistent with an overall deepening of dust mix-ing. Days 6–9 corresponded to the rapid horizontal expansion to-wards planet-encircling scale (Fig. 10). The comparison of theMCS cross sections with the MARCI MDGMs indicates that therewas rapid dust expansion to high altitude after lifting began inNoachis region.

Fig. 11. Log10 of the zonal maximum of the 3 AM MCS dust optical depth for the 10 days shown in Fig. 10.

H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127 125

6. Summary

We have presented dust storm sequences observed in MGSMOC and MRO MARCI Mars Daily Global Maps during Mars years

24–30 to study the origin, evolution and trajectory of large duststorms on Mars. These sequences last for 5 or more sols, travel longdistances and affect multiple regions. Each sequence is composedof daily dust storms that collectively show a continuous develop-

126 H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127

ment history. To simplify, we have neglected dust storms in the po-lar regions and dust storm sequences that may have developedduring global dust storms.

We have visually inspected each MDGM and subjectively iden-tified dust storm sequences. For each sequence, we have recordedthe Ls at the beginning and end, the origination region, the areaoccupied by optically thick dust on each sol, the route traveledand whether or not frontal dust storms are involved.

Our results show that these dust storm sequences are most fre-quently observed during the Ls � 130–350� period, but that a singledefinition of a ‘‘dust storm season’’ combines several different fam-ilies of dust storm types with different seasonal behavior. Specifi-cally, sequences originating from the northern hemisphere(northern sequences) are mainly concentrated in two seasonalwindows during the northern fall (Ls � 180–250�) and winter(Ls � 305–350�). All but two sequences originating from the south-ern hemisphere (southern sequences) are observed duringLs � 135–245� and most of them are concentrated in the periodof Ls � 135–185�. The interval between Ls � 250–305�, which hadtypically been considered to be in the middle of the ‘‘dust stormseason’’ exhibits only one sequence during the study period: the2007 global storm in Mars year 28. The only other global stormin this study is the 2001 global storm (in Mars year 25).

None of the northern sequences are observed to develop intoglobal scale, but some of them develop into the next biggest scale,with peak area >1 � 107 km2 and duration on the order of severalweeks. In comparison, although the explosive growth of the 2001and 2007 global dust storm occur in the southern hemisphere,there appear to be a lack of southern sequences with lifetime ofa few weeks. As a result, northern sequences dominate the duststorm area fraction in non-global dust storm years.

The most preferred origination region is Acidalia, followed byUtopia, Arcadia and Hellas. For non-global dust storms, the maineast–west routes are in the southern hemisphere, but the mainnorth–south routes are from the northern to the southern hemi-sphere. The routes from Acidalia and Utopia show branches inthe southern low/mid-latitudes. They are adopted by a few impres-sive cross-equatorial northern sequences.

Dust storm sequences develop through various combinations ofdevelopment styles – ‘‘consecutive dust storms’’, ‘‘sequential acti-vation’’ or ‘‘merging’’. Dust storm sequences may overlap in time.During our observational period, about 80% of the overlapped se-quences in the northern fall and winter season involve sequenceswith peak area >1 � 107 km2 and are associated with pronouncedincrease in mid-level atmospheric temperatures. In mid/late north-ern summer, there are two southern sequences with peak area>1 � 107 km2, but they are smaller than those during northern falland winter.

The origin of the 2007 global dust storm is somewhat compli-cated by the development of a northern-to-southern hemisphereflushing event from Chryse immediately prior to major dust lift-ing and explosive growth from Noachis. It is unclear whetherthe Chryse storm traveled to Noachis due to image coverage. Fur-ther study is needed to elucidate whether there is a dynamicallink between the Chryse and Noachis dust storm. It is clear fromthe MDGMs that the development of the 2007 global dust stormfrom sub-regional to planet-encircling scale occurred in thesouthern hemisphere. The expansion to high altitudes occurredrapidly after lifting had occurred in Noachis. On this basis, wehave categorized the 2007 global dust storm as a southern hemi-sphere sequence.

MGS MOC and MRO MARCI daily global map provide us anopportunity to derive the statistics and explore the complexity ofthe dust storm sequences that have the potential to grow bigger.Continued long-term spacecraft observation is invaluable forbetter understanding of the martian dust cycle.

Acknowledgments

This paper is supported by NASA’s Mars Data Analysis program.We thank Michael D. Smith for providing the gridded TES and THE-MIS data that are used in Fig. 9. We thank Michael D. Smith andAlexey Pankine for their reviews.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.icarus.2013.10.033.

References

Basu, S., Wilson, R.J., Richardson, M.I., Ingersoll, A.P., 2006. Simulation ofspontaneous and variable global dust storms with the GFDL Mars GCM. J.Geophys. Res. 111, E09004. http://dx.doi.org/10.1029/2005JE002660.

Bell III, J.F. et al., 2009. Mars Reconnaissance Orbiter Mars Color Imager (MARCI):Instrument description, calibration, and performance. J. Geophys. Res. 114, E8.http://dx.doi.org/10.1029/2008JE003315.

Cantor, B.A., 2007. MOC observations of the 2001 Mars planet-encircling dust storm.Icarus 186, 60–96. http://dx.doi.org/10.1016/j.icarus.2006.08.019.

Cantor, B.A., James, P.B., Caplinger, M., Wolff, M.J., 2001. Martian dust storms: 1999Mars Orbiter Camera observations. J. Geophys. Res. 106, 23653–26687.

Hapke, B., 1986. Bidirectional reflectance spectroscopy. 4. Extinction and theopposition effect. Icarus 67, 264–280.

Hartsfeld, N., Ringel, G., 2003. Pearls in Graph Theory: A ComprehensiveIntroduction. Dover, pp. 272.

Hinson, D.P., Wang, H., 2010. Further observations of regional dust storms andbaroclinic eddies in the northern hemisphere of Mars. Icarus 206 (1), 290–305.http://dx.doi.org/10.1016/j.icarus.2009.08.019.

Hinson, D.P., Wang, H., Smith, M.D., 2012. A multi-year survey of dynamics near thesurface in the northern hemisphere of Mars: Short-period baroclinic waves anddust storms. Icarus 219 (1), 307–320. http://dx.doi.org/10.1016/j.icarus.2012.03.001.

Hollingsworth, J.L., Kahre, M.A., 2010. Extratropical cyclones, frontal waves andMars dust: Modeling and considerations. Geophys. Res. Lett. 37, L22202. http://dx.doi.org/10.1029/2010GL044262.

Kahn, R.A., Martin, T.Z., Zurek, R.W., Lee, S.W., 1992. The martian dust cycle. In:Kieffer, H. et al. (Eds.), Mars. Univ. of Ariz. Press, Tucson, pp. 1017–1053.

Kahre, M.A., Murphy, J.R., Haberle, R.M., 2006. Modeling the martian dust cycle andsurface dust reservoirs with the NASA Ames general circulation model. J.Geophys. Res. 111, E6. http://dx.doi.org/10.1029/2005JE002588.

Kleinboehl, A. et al., 2009. Mars Climate Sounder limb profile retrieval ofatmospheric temperature, pressure, and dust and water ice opacity. J.Geophys., Res. 114, E10006. http://dx.doi.org/10.1029/2009JE003358.

Liu, J.J., Richardson, M.I., Wilson, R.J., 2003. An assessment of the global, seasonal,and interannual spacecraft record of martian climate in the thermal infrared. J.Geophys. Res. 108 (E8), 5098. http://dx.doi.org/10.1029/2002JE001921.

Martin, T.Z., 1986. Thermal infrared opacity of the Mars atmosphere. Icarus 66 (1),2–21. http://dx.doi.org/10.1016/0019-1035(86)90003-5.

Martin, T.Z., Richardson, M.I., 1993. New dust opacity mapping from Viking InfraredThermal Mapper data. J. Geophys. Res. 98 (E6), 10941–10949. http://dx.doi.org/10.1029/93JE01044.

Martin, L.J., Zurek, R.W., 1993. An analysis of the history of dust storm activity onMars. J. Geophys. Res. 98, 3221–3246.

Mulholland, D.P., Read, P.L., Lewis, S.R., 2013. Simulating the interannual variability ofmajor dust storms on Mars using variable lifting thresholds. Icarus 223, 344–358.

Newman, C.E., Lewis, S.R., Read, P.L., Forget, F., 2002. Modeling the martian dustcycle. 2. Multiannual radiatively active dust transport simulations. J. Geophys.Res. 107. http://dx.doi.org/10.1029/2002JE001920.

Smith, M.D., 2004. Interannual variability in TES atmospheric observations of Marsduring 1999–2003. Icarus 167, 148–165.

Smith, M.D., 2008. Spacecraft observations of the martian atmosphere. Annu. Rev.Earth Planet. Sci. 36, 191–219. http://dx.doi.org/10.1146/annurev.earth.36.031207.124335.

Smith, M.D., 2009. THEMIS observations of Mars aerosol optical depth from 2002 to2008. Icarus 202, 444–452. http://dx.doi.org/10.1016/j.icarus.2009.03.027.

Smith, M.D., Conrath, R.J., Pearl, J.C., Christensen, P.R., 2002. Thermal EmissionSpectrometer observations of martian planet-encircling dust storm 2001A.Icarus 157 (1), 259–263. http://dx.doi.org/10.1006/icar.2001.6797.

Strausberg, M.J., Wang, H., Richardson, M.I., Ewald, S.P., 2005. Observations of theinitiation and evolution of the 2001 Mars global dust storm. J. Geophys. Res. 110(E2), E02006. http://dx.doi.org/10.1029/2004JE002361.

Toigo, A.D., Richardson, M.I., Wilson, R.J., Wang, H., Ingersoll, A.P., 2002. A first look atdust lifting and dust storms near the south pole of Mars with a mesoscale model. J.Geophys. Res. 107, E7. http://dx.doi.org/10.1029/2001JE001592. Art. No. 5050.

Toigo, A.D., Lee, C., Newman, C.E., Richardson, M.I., 2012. The impact of resolutionon the dynamics of the martian global atmosphere: Varying resolution studies

H. Wang, M.I. Richardson / Icarus 251 (2015) 112–127 127

with the MarsWRF. Icarus 222 (1), 276–288. http://dx.doi.org/10.1016/j.icarus.2012.07.020.

Wang, H., 2007. Dust storms originating in the northern hemisphere during thethird mapping year of Mars Global Surveyor. Icarus 189 (2), 325–343. http://dx.doi.org/10.1016/j.icarus.2007.01.014.

Wang, H., Ingersoll, A.P., 2002. Martian clouds observed by Mars Global SurveyorMars Orbiter Camera. J. Geophys. Res. 107 (E10), 5078. http://dx.doi.org/10.1029/2001JE001815.

Wang, H., Richardson, M.I., Wilson, R.J., Ingersoll, A.P., Toigo, A.D., 2003. Cyclones,tides, and the origin of a cross-equatorial dust storm on Mars.

Wang, H., Zurek, R.W., Richardson, M.I., 2005. Relationship between frontal duststorms and transient eddy activity in the northern hemisphere of Mars as

observed by Mars Global Surveyor. J. Geophys. Res. 110 (E7). http://dx.doi.org/10.1029/2005JE002423.

Wang, H., Toigo, A.D., Richardson, M.I., 2011. Curvilinear features in the southernhemisphere observed by Mars Global Surveyor Mars Orbiter Camera. Icarus 215(1), 242–252. http://dx.doi.org/10.1016/j/icarus.2011.06.029.

Wang, H., Richardson, M.I., Toigo, A.D., Newman, C.E., 2013. Zonal wavenumberthree traveling waves in the northern hemisphere of Mars simulated with ageneral circulation model. Icarus 223 (2), 654–676. http://dx.doi.org/10.1016/j.icarus.2013.01.004.

Wilson, R.J., Hamilton, K., 1996. Comprehensive model simulation of thermal tidesin the martian atmosphere. J. Atmos. Sci. 53 (9), 1290–1326.